We order stock by gut feeling: either shortages or full warehouses
We build a demand forecast per product and per week, accounting for seasonality, promotions and holidays. You see the forecast next to actuals and know where the model is wrong.
Forecast demand, churn and recommendations from your own data
We build machine learning models (programs that find patterns in your sales and customer history). They tell you how much stock to buy, which customers you are about to lose and what to offer each one.
Quote and schedule after your brief
Who it's for
- Online stores and retailers where inventory ties up cash
- Subscription, SaaS and B2B businesses with recurring revenue and churn
- Companies with a catalogue of a few hundred products or more that need recommendations
- Teams with a few years of CRM or accounting data they have never put to use
Problems we solve
Customers leave quietly and we find out too late
A churn model scores each customer on how likely they are to leave. The at-risk list goes into your CRM, and a manager or an AI agent reaches out first.
Everyone sees the same products and conversion is low
We add 'bought together' and 'you may like' recommendations to product pages, emails and messengers. We test the effect on part of your traffic against your normal storefront.
We have data, but we do not know if it is usable
We start with a short data audit. If the history is too short or too messy, we tell you before development starts, not after.
What's included
- Data audit: sources, fields, gaps and duplicates
- Data cleaning and preparation (duplicates, missing values, errors in dates and amounts)
- Success metric and baseline: what exactly the model is compared against, such as your current planning method
- Model development in Python: gradient boosting (CatBoost, LightGBM), time-series models, collaborative filtering for recommendations
- Validation on history the model has not seen, so the accuracy estimate is honest
- Explanations of each prediction: which factors drove the result
- Integration: output lands in your CRM, Google Sheets, store admin or BI dashboard
- Scheduled daily or weekly forecast refresh
- Pilot on real data with a measured KPI
- Documentation and handover: code, models and access in your own accounts
Choose the scope
Use a package to describe what you need. We agree on the quote and schedule after reviewing your brief and requirements.
01
Forecast pilot
You want to test the idea on one problem before a larger investment
- Data audit and KPI definition
- One model: demand or churn
- Validation on historical data
- Report comparing results with your current method
02
Model in production
The pilot proved useful and forecasts now need to flow into daily tools
- Retraining on fresh data
- Integration with CRM, ERP or store
- Automatic scheduled updates
- Dashboard with forecasts and explanations
- Training for your team
03
Recommendations and multi-model
A large catalogue or several problems at once: demand, churn, personalised picks
- Recommendation engine for site and email
- Two or three models in one pipeline
- A/B test (comparison on two groups of visitors)
- Connection to an AI agent or workflow automation
- Quality monitoring after launch
How we work
Brief and requirements
You describe the task and current workflow. We clarify requirements, choose a first stage and prepare a quote.
Data audit
We review your sales, customer or event history and fix the KPI. We tell you plainly whether there is enough data.
Pilot
We build the first model and test it on data it has not seen. We compare it with how you plan today.
Integration
We connect the forecast to your CRM, ERP or site. The output has to land where your people already work.
Launch and monitoring
We watch quality, because data and customer behaviour drift. Code, models and access belong to you under the agreed terms.
Scope and limits
- We do not promise perfect forecasts. A model reduces uncertainty but does not remove it, and it cannot predict sudden shocks such as supply disruptions or regulation changes.
- We do not take on a project when the data is too thin. Demand and churn models usually need a history of several seasons; otherwise it is better to start by collecting data.
- We do not sell AI for its own sake. If a spreadsheet or a simple rule solves the problem, we will say so.
FAQ
How much data do I need?
We prepare a quote after reviewing your brief and requirements. A first stage can cover: data audit and KPI definition; One model: demand or churn. Scope, payment and schedule are agreed before work starts.
Is this the same as ChatGPT?
No. Large language models work with text. Forecasting numbers and behaviour is usually better handled by classical models on tabular data. We add a language model where it helps, for example to explain a forecast in plain words.
How do you measure success?
We agree on a KPI with you before starting. For example, how much more accurate the forecast is than your current plan, or how many at-risk customers you win back. Without a baseline, results cannot be judged.
Who owns the model and the data?
We agree on the handover of code, materials and account access before work starts. Third-party licences and services are listed separately when needed.
Where is my data stored?
In your own storage or cloud account. We work with the access you grant and, where needed, remove personal data before training the model.
How is the project paid for?
In stages: audit and pilot, then integration, then launch. If the pilot does not show value, you do not pay for the later stages.
Start with your task
Describe your business, the outcome you need and your current workflow. We will clarify the requirements and prepare a quote after your brief.